* fix fruit dataset

* fix md5

* change voc_loader

* refine doc for transfer learning

* add face doc
* update face detection readme and add demo jpg

* refine download sh to py

* update QUICK_STARTED

* add multi-scale test

* add class aware sampling strategy
* remove redundancy code

This reverts commit b591cc13.

* Add face detection configs and main code.
* Integration multi-scale evaluation.

* unify reader to dataloader

* add check_version

Update ppdet/utils/checkpoint.py to skip the var shape check when the var is not initialized. (#3452)

* Support finetune by custom dataset

* add finetune args

* add load finetune

* reconstruct load

* add transfer learning doc

* add fruit demo

* add quick start

* add data preprocessing FAQ

* Initial support for distributed training

* Housekeep on rank 0 only

* Conform to models convention

* refine auto download logs

* Add benchmark (fps) in MODEL_ZOO doc

* fix train on VOC on multi-GPU hang

* not set persistable in parse_fetches

* add gn doc

# Add VGG-SSD on VOC and COCO dataset.
# Add config and model zoo. 
# Refine bbox2out and draw_bbox:
    - Add bbox de-normalize in bbox2out function.
    - Remove bbox de-normalize in draw_bbox.

* refine weight download

* fix comment

* add save best model & fix dcn configs

* fix yolov3 voc eval in train

* fix comment

* fix easy problem

Add PaddleDetection docs & some model configs

* add voc_eval and yolo_darknet_voc

* add yolov3_darknet_voc in MODEL_ZOO

* fix default im_size

* fix MODEL_ZOO note

* fix is_bbox_normalized

* extract map to map_utils.py

* update yolov3_dorknet_voc mixup

* add yolov3_r34_voc

* add yolov3_mobilenet_v1_voc

* fix drop empty in VAL mode

* use cfg.num_classes

* assert metric valid

* enable difficulty can be None

* add comment for bbox_eval

* num_classes in retinanet

option should not be overridden by subsequent options of the same module.